Apple has released public code for LiTo, an ICLR 2026 image-to-3D model that preserves lighting, reflections, and material behavior, signaling a shift toward usable 3D AI tooling for developers in spatial computing, games, and enterprise AR.

The data strategy uses Objaverse and ObjaverseXL datasets, totaling about 240,100 samples after filtering, to enable reproducibility and benchmarking for researchers.

Apple emphasizes practical tooling rather than ready-for-production products, noting that full training and experimentation require strong GPU access, making LiTo ideal for prototyping, benchmarking, and pipeline research.

LiTo sits among a crowded field of open 3D generation efforts, including Microsoft TRELLIS, Tencent Hunyuan3D, and Roblox Cube 3D, but targets higher fidelity, better alignment, and lighting consistency, addressing orientation and camera coordinate issues in competitors.

LiTo represents a serious move from research demos toward infrastructure that could become part of developers’ standard pipelines for 3D content creation.

LiTo couples geometry and view-dependent appearance in a single 3D representation, enabling more faithful rendering of shiny and glossy objects by modeling light interaction from multiple angles.

The release Includes code, demos, data splits, configuration files, and licensing details, with public updates to GitHub as of May 23, 2026, and training performance noted at about 4.7 seconds on Nvidia H100 after compilation.

Although not framed as part of Vision Pro, the LiTo release supports developers building spatial computing assets, simulations, digital twins, and immersive apps, potentially speeding workflows from draft to deployable content.